fs/xfs/xfs_super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   4  * All Rights Reserved.
   5  */
   6
   7 #include "xfs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_sb.h"
  13 #include "xfs_mount.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_btree.h"
  16 #include "xfs_bmap.h"
  17 #include "xfs_alloc.h"
  18 #include "xfs_fsops.h"
  19 #include "xfs_trans.h"
  20 #include "xfs_buf_item.h"
  21 #include "xfs_log.h"
  22 #include "xfs_log_priv.h"
  23 #include "xfs_dir2.h"
  24 #include "xfs_extfree_item.h"
  25 #include "xfs_mru_cache.h"
  26 #include "xfs_inode_item.h"
  27 #include "xfs_icache.h"
  28 #include "xfs_trace.h"
  29 #include "xfs_icreate_item.h"
  30 #include "xfs_filestream.h"
  31 #include "xfs_quota.h"
  32 #include "xfs_sysfs.h"
  33 #include "xfs_ondisk.h"
  34 #include "xfs_rmap_item.h"
  35 #include "xfs_refcount_item.h"
  36 #include "xfs_bmap_item.h"
  37 #include "xfs_reflink.h"
  38 #include "xfs_pwork.h"
  39 #include "xfs_ag.h"
  40
  41 #include <linux/magic.h>
  42 #include <linux/fs_context.h>
  43 #include <linux/fs_parser.h>
  44
  45 static const struct super_operations xfs_super_operations;
  46
  47 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
  48 #ifdef DEBUG
  49 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
  50 #endif
  51
  52 enum xfs_dax_mode {
  53         XFS_DAX_INODE = 0,
  54         XFS_DAX_ALWAYS = 1,
  55         XFS_DAX_NEVER = 2,
  56 };
  57
  58 static void
  59 xfs_mount_set_dax_mode(
  60         struct xfs_mount        *mp,
  61         enum xfs_dax_mode       mode)
  62 {
  63         switch (mode) {
  64         case XFS_DAX_INODE:
  65                 mp->m_flags &= ~(XFS_MOUNT_DAX_ALWAYS | XFS_MOUNT_DAX_NEVER);
  66                 break;
  67         case XFS_DAX_ALWAYS:
  68                 mp->m_flags |= XFS_MOUNT_DAX_ALWAYS;
  69                 mp->m_flags &= ~XFS_MOUNT_DAX_NEVER;
  70                 break;
  71         case XFS_DAX_NEVER:
  72                 mp->m_flags |= XFS_MOUNT_DAX_NEVER;
  73                 mp->m_flags &= ~XFS_MOUNT_DAX_ALWAYS;
  74                 break;
  75         }
  76 }
  77
  78 static const struct constant_table dax_param_enums[] = {
  79         {"inode",       XFS_DAX_INODE },
  80         {"always",      XFS_DAX_ALWAYS },
  81         {"never",       XFS_DAX_NEVER },
  82         {}
  83 };
  84
  85 /*
  86  * Table driven mount option parser.
  87  */
  88 enum {
  89         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
  90         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
  91         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
  92         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
  93         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
  94         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
  95         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
  96         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
  97         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
  98 };
  99
 100 static const struct fs_parameter_spec xfs_fs_parameters[] = {
 101         fsparam_u32("logbufs",          Opt_logbufs),
 102         fsparam_string("logbsize",      Opt_logbsize),
 103         fsparam_string("logdev",        Opt_logdev),
 104         fsparam_string("rtdev",         Opt_rtdev),
 105         fsparam_flag("wsync",           Opt_wsync),
 106         fsparam_flag("noalign",         Opt_noalign),
 107         fsparam_flag("swalloc",         Opt_swalloc),
 108         fsparam_u32("sunit",            Opt_sunit),
 109         fsparam_u32("swidth",           Opt_swidth),
 110         fsparam_flag("nouuid",          Opt_nouuid),
 111         fsparam_flag("grpid",           Opt_grpid),
 112         fsparam_flag("nogrpid",         Opt_nogrpid),
 113         fsparam_flag("bsdgroups",       Opt_bsdgroups),
 114         fsparam_flag("sysvgroups",      Opt_sysvgroups),
 115         fsparam_string("allocsize",     Opt_allocsize),
 116         fsparam_flag("norecovery",      Opt_norecovery),
 117         fsparam_flag("inode64",         Opt_inode64),
 118         fsparam_flag("inode32",         Opt_inode32),
 119         fsparam_flag("ikeep",           Opt_ikeep),
 120         fsparam_flag("noikeep",         Opt_noikeep),
 121         fsparam_flag("largeio",         Opt_largeio),
 122         fsparam_flag("nolargeio",       Opt_nolargeio),
 123         fsparam_flag("attr2",           Opt_attr2),
 124         fsparam_flag("noattr2",         Opt_noattr2),
 125         fsparam_flag("filestreams",     Opt_filestreams),
 126         fsparam_flag("quota",           Opt_quota),
 127         fsparam_flag("noquota",         Opt_noquota),
 128         fsparam_flag("usrquota",        Opt_usrquota),
 129         fsparam_flag("grpquota",        Opt_grpquota),
 130         fsparam_flag("prjquota",        Opt_prjquota),
 131         fsparam_flag("uquota",          Opt_uquota),
 132         fsparam_flag("gquota",          Opt_gquota),
 133         fsparam_flag("pquota",          Opt_pquota),
 134         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
 135         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
 136         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
 137         fsparam_flag("qnoenforce",      Opt_qnoenforce),
 138         fsparam_flag("discard",         Opt_discard),
 139         fsparam_flag("nodiscard",       Opt_nodiscard),
 140         fsparam_flag("dax",             Opt_dax),
 141         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
 142         {}
 143 };
 144
 145 struct proc_xfs_info {
 146         uint64_t        flag;
 147         char            *str;
 148 };
 149
 150 static int
 151 xfs_fs_show_options(
 152         struct seq_file         *m,
 153         struct dentry           *root)
 154 {
 155         static struct proc_xfs_info xfs_info_set[] = {
 156                 /* the few simple ones we can get from the mount struct */
 157                 { XFS_MOUNT_IKEEP,              ",ikeep" },
 158                 { XFS_MOUNT_WSYNC,              ",wsync" },
 159                 { XFS_MOUNT_NOALIGN,            ",noalign" },
 160                 { XFS_MOUNT_SWALLOC,            ",swalloc" },
 161                 { XFS_MOUNT_NOUUID,             ",nouuid" },
 162                 { XFS_MOUNT_NORECOVERY,         ",norecovery" },
 163                 { XFS_MOUNT_ATTR2,              ",attr2" },
 164                 { XFS_MOUNT_FILESTREAMS,        ",filestreams" },
 165                 { XFS_MOUNT_GRPID,              ",grpid" },
 166                 { XFS_MOUNT_DISCARD,            ",discard" },
 167                 { XFS_MOUNT_LARGEIO,            ",largeio" },
 168                 { XFS_MOUNT_DAX_ALWAYS,         ",dax=always" },
 169                 { XFS_MOUNT_DAX_NEVER,          ",dax=never" },
 170                 { 0, NULL }
 171         };
 172         struct xfs_mount        *mp = XFS_M(root->d_sb);
 173         struct proc_xfs_info    *xfs_infop;
 174
 175         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 176                 if (mp->m_flags & xfs_infop->flag)
 177                         seq_puts(m, xfs_infop->str);
 178         }
 179
 180         seq_printf(m, ",inode%d",
 181                 (mp->m_flags & XFS_MOUNT_SMALL_INUMS) ? 32 : 64);
 182
 183         if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
 184                 seq_printf(m, ",allocsize=%dk",
 185                            (1 << mp->m_allocsize_log) >> 10);
 186
 187         if (mp->m_logbufs > 0)
 188                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 189         if (mp->m_logbsize > 0)
 190                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 191
 192         if (mp->m_logname)
 193                 seq_show_option(m, "logdev", mp->m_logname);
 194         if (mp->m_rtname)
 195                 seq_show_option(m, "rtdev", mp->m_rtname);
 196
 197         if (mp->m_dalign > 0)
 198                 seq_printf(m, ",sunit=%d",
 199                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 200         if (mp->m_swidth > 0)
 201                 seq_printf(m, ",swidth=%d",
 202                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 203
 204         if (mp->m_qflags & XFS_UQUOTA_ACCT) {
 205                 if (mp->m_qflags & XFS_UQUOTA_ENFD)
 206                         seq_puts(m, ",usrquota");
 207                 else
 208                         seq_puts(m, ",uqnoenforce");
 209         }
 210
 211         if (mp->m_qflags & XFS_PQUOTA_ACCT) {
 212                 if (mp->m_qflags & XFS_PQUOTA_ENFD)
 213                         seq_puts(m, ",prjquota");
 214                 else
 215                         seq_puts(m, ",pqnoenforce");
 216         }
 217         if (mp->m_qflags & XFS_GQUOTA_ACCT) {
 218                 if (mp->m_qflags & XFS_GQUOTA_ENFD)
 219                         seq_puts(m, ",grpquota");
 220                 else
 221                         seq_puts(m, ",gqnoenforce");
 222         }
 223
 224         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 225                 seq_puts(m, ",noquota");
 226
 227         return 0;
 228 }
 229
 230 /*
 231  * Set parameters for inode allocation heuristics, taking into account
 232  * filesystem size and inode32/inode64 mount options; i.e. specifically
 233  * whether or not XFS_MOUNT_SMALL_INUMS is set.
 234  *
 235  * Inode allocation patterns are altered only if inode32 is requested
 236  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
 237  * If altered, XFS_MOUNT_32BITINODES is set as well.
 238  *
 239  * An agcount independent of that in the mount structure is provided
 240  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 241  * to the potentially higher ag count.
 242  *
 243  * Returns the maximum AG index which may contain inodes.
 244  */
 245 xfs_agnumber_t
 246 xfs_set_inode_alloc(
 247         struct xfs_mount *mp,
 248         xfs_agnumber_t  agcount)
 249 {
 250         xfs_agnumber_t  index;
 251         xfs_agnumber_t  maxagi = 0;
 252         xfs_sb_t        *sbp = &mp->m_sb;
 253         xfs_agnumber_t  max_metadata;
 254         xfs_agino_t     agino;
 255         xfs_ino_t       ino;
 256
 257         /*
 258          * Calculate how much should be reserved for inodes to meet
 259          * the max inode percentage.  Used only for inode32.
 260          */
 261         if (M_IGEO(mp)->maxicount) {
 262                 uint64_t        icount;
 263
 264                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 265                 do_div(icount, 100);
 266                 icount += sbp->sb_agblocks - 1;
 267                 do_div(icount, sbp->sb_agblocks);
 268                 max_metadata = icount;
 269         } else {
 270                 max_metadata = agcount;
 271         }
 272
 273         /* Get the last possible inode in the filesystem */
 274         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 275         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 276
 277         /*
 278          * If user asked for no more than 32-bit inodes, and the fs is
 279          * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
 280          * the allocator to accommodate the request.
 281          */
 282         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
 283                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 284         else
 285                 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
 286
 287         for (index = 0; index < agcount; index++) {
 288                 struct xfs_perag        *pag;
 289
 290                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 291
 292                 pag = xfs_perag_get(mp, index);
 293
 294                 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
 295                         if (ino > XFS_MAXINUMBER_32) {
 296                                 pag->pagi_inodeok = 0;
 297                                 pag->pagf_metadata = 0;
 298                         } else {
 299                                 pag->pagi_inodeok = 1;
 300                                 maxagi++;
 301                                 if (index < max_metadata)
 302                                         pag->pagf_metadata = 1;
 303                                 else
 304                                         pag->pagf_metadata = 0;
 305                         }
 306                 } else {
 307                         pag->pagi_inodeok = 1;
 308                         pag->pagf_metadata = 0;
 309                 }
 310
 311                 xfs_perag_put(pag);
 312         }
 313
 314         return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
 315 }
 316
 317 STATIC int
 318 xfs_blkdev_get(
 319         xfs_mount_t             *mp,
 320         const char              *name,
 321         struct block_device     **bdevp)
 322 {
 323         int                     error = 0;
 324
 325         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 326                                     mp);
 327         if (IS_ERR(*bdevp)) {
 328                 error = PTR_ERR(*bdevp);
 329                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 330         }
 331
 332         return error;
 333 }
 334
 335 STATIC void
 336 xfs_blkdev_put(
 337         struct block_device     *bdev)
 338 {
 339         if (bdev)
 340                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 341 }
 342
 343 STATIC void
 344 xfs_close_devices(
 345         struct xfs_mount        *mp)
 346 {
 347         struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
 348
 349         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 350                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 351                 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
 352
 353                 xfs_free_buftarg(mp->m_logdev_targp);
 354                 xfs_blkdev_put(logdev);
 355                 fs_put_dax(dax_logdev);
 356         }
 357         if (mp->m_rtdev_targp) {
 358                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 359                 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
 360
 361                 xfs_free_buftarg(mp->m_rtdev_targp);
 362                 xfs_blkdev_put(rtdev);
 363                 fs_put_dax(dax_rtdev);
 364         }
 365         xfs_free_buftarg(mp->m_ddev_targp);
 366         fs_put_dax(dax_ddev);
 367 }
 368
 369 /*
 370  * The file system configurations are:
 371  *      (1) device (partition) with data and internal log
 372  *      (2) logical volume with data and log subvolumes.
 373  *      (3) logical volume with data, log, and realtime subvolumes.
 374  *
 375  * We only have to handle opening the log and realtime volumes here if
 376  * they are present.  The data subvolume has already been opened by
 377  * get_sb_bdev() and is stored in sb->s_bdev.
 378  */
 379 STATIC int
 380 xfs_open_devices(
 381         struct xfs_mount        *mp)
 382 {
 383         struct block_device     *ddev = mp->m_super->s_bdev;
 384         struct dax_device       *dax_ddev = fs_dax_get_by_bdev(ddev);
 385         struct dax_device       *dax_logdev = NULL, *dax_rtdev = NULL;
 386         struct block_device     *logdev = NULL, *rtdev = NULL;
 387         int                     error;
 388
 389         /*
 390          * Open real time and log devices - order is important.
 391          */
 392         if (mp->m_logname) {
 393                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 394                 if (error)
 395                         goto out;
 396                 dax_logdev = fs_dax_get_by_bdev(logdev);
 397         }
 398
 399         if (mp->m_rtname) {
 400                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 401                 if (error)
 402                         goto out_close_logdev;
 403
 404                 if (rtdev == ddev || rtdev == logdev) {
 405                         xfs_warn(mp,
 406         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 407                         error = -EINVAL;
 408                         goto out_close_rtdev;
 409                 }
 410                 dax_rtdev = fs_dax_get_by_bdev(rtdev);
 411         }
 412
 413         /*
 414          * Setup xfs_mount buffer target pointers
 415          */
 416         error = -ENOMEM;
 417         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
 418         if (!mp->m_ddev_targp)
 419                 goto out_close_rtdev;
 420
 421         if (rtdev) {
 422                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
 423                 if (!mp->m_rtdev_targp)
 424                         goto out_free_ddev_targ;
 425         }
 426
 427         if (logdev && logdev != ddev) {
 428                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
 429                 if (!mp->m_logdev_targp)
 430                         goto out_free_rtdev_targ;
 431         } else {
 432                 mp->m_logdev_targp = mp->m_ddev_targp;
 433         }
 434
 435         return 0;
 436
 437  out_free_rtdev_targ:
 438         if (mp->m_rtdev_targp)
 439                 xfs_free_buftarg(mp->m_rtdev_targp);
 440  out_free_ddev_targ:
 441         xfs_free_buftarg(mp->m_ddev_targp);
 442  out_close_rtdev:
 443         xfs_blkdev_put(rtdev);
 444         fs_put_dax(dax_rtdev);
 445  out_close_logdev:
 446         if (logdev && logdev != ddev) {
 447                 xfs_blkdev_put(logdev);
 448                 fs_put_dax(dax_logdev);
 449         }
 450  out:
 451         fs_put_dax(dax_ddev);
 452         return error;
 453 }
 454
 455 /*
 456  * Setup xfs_mount buffer target pointers based on superblock
 457  */
 458 STATIC int
 459 xfs_setup_devices(
 460         struct xfs_mount        *mp)
 461 {
 462         int                     error;
 463
 464         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 465         if (error)
 466                 return error;
 467
 468         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 469                 unsigned int    log_sector_size = BBSIZE;
 470
 471                 if (xfs_sb_version_hassector(&mp->m_sb))
 472                         log_sector_size = mp->m_sb.sb_logsectsize;
 473                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 474                                             log_sector_size);
 475                 if (error)
 476                         return error;
 477         }
 478         if (mp->m_rtdev_targp) {
 479                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 480                                             mp->m_sb.sb_sectsize);
 481                 if (error)
 482                         return error;
 483         }
 484
 485         return 0;
 486 }
 487
 488 STATIC int
 489 xfs_init_mount_workqueues(
 490         struct xfs_mount        *mp)
 491 {
 492         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 493                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 494                         1, mp->m_super->s_id);
 495         if (!mp->m_buf_workqueue)
 496                 goto out;
 497
 498         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 499                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 500                         0, mp->m_super->s_id);
 501         if (!mp->m_unwritten_workqueue)
 502                 goto out_destroy_buf;
 503
 504         mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
 505                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM | WQ_UNBOUND),
 506                         0, mp->m_super->s_id);
 507         if (!mp->m_cil_workqueue)
 508                 goto out_destroy_unwritten;
 509
 510         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 511                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 512                         0, mp->m_super->s_id);
 513         if (!mp->m_reclaim_workqueue)
 514                 goto out_destroy_cil;
 515
 516         mp->m_gc_workqueue = alloc_workqueue("xfs-gc/%s",
 517                         WQ_SYSFS | WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM,
 518                         0, mp->m_super->s_id);
 519         if (!mp->m_gc_workqueue)
 520                 goto out_destroy_reclaim;
 521
 522         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
 523                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
 524         if (!mp->m_sync_workqueue)
 525                 goto out_destroy_eofb;
 526
 527         return 0;
 528
 529 out_destroy_eofb:
 530         destroy_workqueue(mp->m_gc_workqueue);
 531 out_destroy_reclaim:
 532         destroy_workqueue(mp->m_reclaim_workqueue);
 533 out_destroy_cil:
 534         destroy_workqueue(mp->m_cil_workqueue);
 535 out_destroy_unwritten:
 536         destroy_workqueue(mp->m_unwritten_workqueue);
 537 out_destroy_buf:
 538         destroy_workqueue(mp->m_buf_workqueue);
 539 out:
 540         return -ENOMEM;
 541 }
 542
 543 STATIC void
 544 xfs_destroy_mount_workqueues(
 545         struct xfs_mount        *mp)
 546 {
 547         destroy_workqueue(mp->m_sync_workqueue);
 548         destroy_workqueue(mp->m_gc_workqueue);
 549         destroy_workqueue(mp->m_reclaim_workqueue);
 550         destroy_workqueue(mp->m_cil_workqueue);
 551         destroy_workqueue(mp->m_unwritten_workqueue);
 552         destroy_workqueue(mp->m_buf_workqueue);
 553 }
 554
 555 static void
 556 xfs_flush_inodes_worker(
 557         struct work_struct      *work)
 558 {
 559         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
 560                                                    m_flush_inodes_work);
 561         struct super_block      *sb = mp->m_super;
 562
 563         if (down_read_trylock(&sb->s_umount)) {
 564                 sync_inodes_sb(sb);
 565                 up_read(&sb->s_umount);
 566         }
 567 }
 568
 569 /*
 570  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 571  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 572  * for IO to complete so that we effectively throttle multiple callers to the
 573  * rate at which IO is completing.
 574  */
 575 void
 576 xfs_flush_inodes(
 577         struct xfs_mount        *mp)
 578 {
 579         /*
 580          * If flush_work() returns true then that means we waited for a flush
 581          * which was already in progress.  Don't bother running another scan.
 582          */
 583         if (flush_work(&mp->m_flush_inodes_work))
 584                 return;
 585
 586         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 587         flush_work(&mp->m_flush_inodes_work);
 588 }
 589
 590 /* Catch misguided souls that try to use this interface on XFS */
 591 STATIC struct inode *
 592 xfs_fs_alloc_inode(
 593         struct super_block      *sb)
 594 {
 595         BUG();
 596         return NULL;
 597 }
 598
 599 #ifdef DEBUG
 600 static void
 601 xfs_check_delalloc(
 602         struct xfs_inode        *ip,
 603         int                     whichfork)
 604 {
 605         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
 606         struct xfs_bmbt_irec    got;
 607         struct xfs_iext_cursor  icur;
 608
 609         if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
 610                 return;
 611         do {
 612                 if (isnullstartblock(got.br_startblock)) {
 613                         xfs_warn(ip->i_mount,
 614         "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
 615                                 ip->i_ino,
 616                                 whichfork == XFS_DATA_FORK ? "data" : "cow",
 617                                 got.br_startoff, got.br_blockcount);
 618                 }
 619         } while (xfs_iext_next_extent(ifp, &icur, &got));
 620 }
 621 #else
 622 #define xfs_check_delalloc(ip, whichfork)       do { } while (0)
 623 #endif
 624
 625 /*
 626  * Now that the generic code is guaranteed not to be accessing
 627  * the linux inode, we can inactivate and reclaim the inode.
 628  */
 629 STATIC void
 630 xfs_fs_destroy_inode(
 631         struct inode            *inode)
 632 {
 633         struct xfs_inode        *ip = XFS_I(inode);
 634
 635         trace_xfs_destroy_inode(ip);
 636
 637         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 638         XFS_STATS_INC(ip->i_mount, vn_rele);
 639         XFS_STATS_INC(ip->i_mount, vn_remove);
 640
 641         xfs_inactive(ip);
 642
 643         if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
 644                 xfs_check_delalloc(ip, XFS_DATA_FORK);
 645                 xfs_check_delalloc(ip, XFS_COW_FORK);
 646                 ASSERT(0);
 647         }
 648
 649         XFS_STATS_INC(ip->i_mount, vn_reclaim);
 650
 651         /*
 652          * We should never get here with one of the reclaim flags already set.
 653          */
 654         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
 655         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
 656
 657         /*
 658          * We always use background reclaim here because even if the inode is
 659          * clean, it still may be under IO and hence we have wait for IO
 660          * completion to occur before we can reclaim the inode. The background
 661          * reclaim path handles this more efficiently than we can here, so
 662          * simply let background reclaim tear down all inodes.
 663          */
 664         xfs_inode_mark_reclaimable(ip);
 665 }
 666
 667 static void
 668 xfs_fs_dirty_inode(
 669         struct inode                    *inode,
 670         int                             flag)
 671 {
 672         struct xfs_inode                *ip = XFS_I(inode);
 673         struct xfs_mount                *mp = ip->i_mount;
 674         struct xfs_trans                *tp;
 675
 676         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 677                 return;
 678         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
 679                 return;
 680
 681         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 682                 return;
 683         xfs_ilock(ip, XFS_ILOCK_EXCL);
 684         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 685         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 686         xfs_trans_commit(tp);
 687 }
 688
 689 /*
 690  * Slab object creation initialisation for the XFS inode.
 691  * This covers only the idempotent fields in the XFS inode;
 692  * all other fields need to be initialised on allocation
 693  * from the slab. This avoids the need to repeatedly initialise
 694  * fields in the xfs inode that left in the initialise state
 695  * when freeing the inode.
 696  */
 697 STATIC void
 698 xfs_fs_inode_init_once(
 699         void                    *inode)
 700 {
 701         struct xfs_inode        *ip = inode;
 702
 703         memset(ip, 0, sizeof(struct xfs_inode));
 704
 705         /* vfs inode */
 706         inode_init_once(VFS_I(ip));
 707
 708         /* xfs inode */
 709         atomic_set(&ip->i_pincount, 0);
 710         spin_lock_init(&ip->i_flags_lock);
 711
 712         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 713                      "xfsino", ip->i_ino);
 714 }
 715
 716 /*
 717  * We do an unlocked check for XFS_IDONTCACHE here because we are already
 718  * serialised against cache hits here via the inode->i_lock and igrab() in
 719  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 720  * racing with us, and it avoids needing to grab a spinlock here for every inode
 721  * we drop the final reference on.
 722  */
 723 STATIC int
 724 xfs_fs_drop_inode(
 725         struct inode            *inode)
 726 {
 727         struct xfs_inode        *ip = XFS_I(inode);
 728
 729         /*
 730          * If this unlinked inode is in the middle of recovery, don't
 731          * drop the inode just yet; log recovery will take care of
 732          * that.  See the comment for this inode flag.
 733          */
 734         if (ip->i_flags & XFS_IRECOVERY) {
 735                 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
 736                 return 0;
 737         }
 738
 739         return generic_drop_inode(inode);
 740 }
 741
 742 static void
 743 xfs_mount_free(
 744         struct xfs_mount        *mp)
 745 {
 746         kfree(mp->m_rtname);
 747         kfree(mp->m_logname);
 748         kmem_free(mp);
 749 }
 750
 751 STATIC int
 752 xfs_fs_sync_fs(
 753         struct super_block      *sb,
 754         int                     wait)
 755 {
 756         struct xfs_mount        *mp = XFS_M(sb);
 757
 758         /*
 759          * Doing anything during the async pass would be counterproductive.
 760          */
 761         if (!wait)
 762                 return 0;
 763
 764         xfs_log_force(mp, XFS_LOG_SYNC);
 765         if (laptop_mode) {
 766                 /*
 767                  * The disk must be active because we're syncing.
 768                  * We schedule log work now (now that the disk is
 769                  * active) instead of later (when it might not be).
 770                  */
 771                 flush_delayed_work(&mp->m_log->l_work);
 772         }
 773
 774         return 0;
 775 }
 776
 777 STATIC int
 778 xfs_fs_statfs(
 779         struct dentry           *dentry,
 780         struct kstatfs          *statp)
 781 {
 782         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
 783         xfs_sb_t                *sbp = &mp->m_sb;
 784         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
 785         uint64_t                fakeinos, id;
 786         uint64_t                icount;
 787         uint64_t                ifree;
 788         uint64_t                fdblocks;
 789         xfs_extlen_t            lsize;
 790         int64_t                 ffree;
 791
 792         statp->f_type = XFS_SUPER_MAGIC;
 793         statp->f_namelen = MAXNAMELEN - 1;
 794
 795         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 796         statp->f_fsid = u64_to_fsid(id);
 797
 798         icount = percpu_counter_sum(&mp->m_icount);
 799         ifree = percpu_counter_sum(&mp->m_ifree);
 800         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 801
 802         spin_lock(&mp->m_sb_lock);
 803         statp->f_bsize = sbp->sb_blocksize;
 804         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 805         statp->f_blocks = sbp->sb_dblocks - lsize;
 806         spin_unlock(&mp->m_sb_lock);
 807
 808         /* make sure statp->f_bfree does not underflow */
 809         statp->f_bfree = max_t(int64_t, fdblocks - mp->m_alloc_set_aside, 0);
 810         statp->f_bavail = statp->f_bfree;
 811
 812         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 813         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 814         if (M_IGEO(mp)->maxicount)
 815                 statp->f_files = min_t(typeof(statp->f_files),
 816                                         statp->f_files,
 817                                         M_IGEO(mp)->maxicount);
 818
 819         /* If sb_icount overshot maxicount, report actual allocation */
 820         statp->f_files = max_t(typeof(statp->f_files),
 821                                         statp->f_files,
 822                                         sbp->sb_icount);
 823
 824         /* make sure statp->f_ffree does not underflow */
 825         ffree = statp->f_files - (icount - ifree);
 826         statp->f_ffree = max_t(int64_t, ffree, 0);
 827
 828
 829         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 830             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 831                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 832                 xfs_qm_statvfs(ip, statp);
 833
 834         if (XFS_IS_REALTIME_MOUNT(mp) &&
 835             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 836                 statp->f_blocks = sbp->sb_rblocks;
 837                 statp->f_bavail = statp->f_bfree =
 838                         sbp->sb_frextents * sbp->sb_rextsize;
 839         }
 840
 841         return 0;
 842 }
 843
 844 STATIC void
 845 xfs_save_resvblks(struct xfs_mount *mp)
 846 {
 847         uint64_t resblks = 0;
 848
 849         mp->m_resblks_save = mp->m_resblks;
 850         xfs_reserve_blocks(mp, &resblks, NULL);
 851 }
 852
 853 STATIC void
 854 xfs_restore_resvblks(struct xfs_mount *mp)
 855 {
 856         uint64_t resblks;
 857
 858         if (mp->m_resblks_save) {
 859                 resblks = mp->m_resblks_save;
 860                 mp->m_resblks_save = 0;
 861         } else
 862                 resblks = xfs_default_resblks(mp);
 863
 864         xfs_reserve_blocks(mp, &resblks, NULL);
 865 }
 866
 867 /*
 868  * Second stage of a freeze. The data is already frozen so we only
 869  * need to take care of the metadata. Once that's done sync the superblock
 870  * to the log to dirty it in case of a crash while frozen. This ensures that we
 871  * will recover the unlinked inode lists on the next mount.
 872  */
 873 STATIC int
 874 xfs_fs_freeze(
 875         struct super_block      *sb)
 876 {
 877         struct xfs_mount        *mp = XFS_M(sb);
 878         unsigned int            flags;
 879         int                     ret;
 880
 881         /*
 882          * The filesystem is now frozen far enough that memory reclaim
 883          * cannot safely operate on the filesystem. Hence we need to
 884          * set a GFP_NOFS context here to avoid recursion deadlocks.
 885          */
 886         flags = memalloc_nofs_save();
 887         xfs_blockgc_stop(mp);
 888         xfs_save_resvblks(mp);
 889         ret = xfs_log_quiesce(mp);
 890         memalloc_nofs_restore(flags);
 891         return ret;
 892 }
 893
 894 STATIC int
 895 xfs_fs_unfreeze(
 896         struct super_block      *sb)
 897 {
 898         struct xfs_mount        *mp = XFS_M(sb);
 899
 900         xfs_restore_resvblks(mp);
 901         xfs_log_work_queue(mp);
 902         xfs_blockgc_start(mp);
 903         return 0;
 904 }
 905
 906 /*
 907  * This function fills in xfs_mount_t fields based on mount args.
 908  * Note: the superblock _has_ now been read in.
 909  */
 910 STATIC int
 911 xfs_finish_flags(
 912         struct xfs_mount        *mp)
 913 {
 914         int                     ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
 915
 916         /* Fail a mount where the logbuf is smaller than the log stripe */
 917         if (xfs_sb_version_haslogv2(&mp->m_sb)) {
 918                 if (mp->m_logbsize <= 0 &&
 919                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 920                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 921                 } else if (mp->m_logbsize > 0 &&
 922                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
 923                         xfs_warn(mp,
 924                 "logbuf size must be greater than or equal to log stripe size");
 925                         return -EINVAL;
 926                 }
 927         } else {
 928                 /* Fail a mount if the logbuf is larger than 32K */
 929                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 930                         xfs_warn(mp,
 931                 "logbuf size for version 1 logs must be 16K or 32K");
 932                         return -EINVAL;
 933                 }
 934         }
 935
 936         /*
 937          * V5 filesystems always use attr2 format for attributes.
 938          */
 939         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 940             (mp->m_flags & XFS_MOUNT_NOATTR2)) {
 941                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
 942                              "attr2 is always enabled for V5 filesystems.");
 943                 return -EINVAL;
 944         }
 945
 946         /*
 947          * mkfs'ed attr2 will turn on attr2 mount unless explicitly
 948          * told by noattr2 to turn it off
 949          */
 950         if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 951             !(mp->m_flags & XFS_MOUNT_NOATTR2))
 952                 mp->m_flags |= XFS_MOUNT_ATTR2;
 953
 954         /*
 955          * prohibit r/w mounts of read-only filesystems
 956          */
 957         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 958                 xfs_warn(mp,
 959                         "cannot mount a read-only filesystem as read-write");
 960                 return -EROFS;
 961         }
 962
 963         if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
 964             (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
 965             !xfs_sb_version_has_pquotino(&mp->m_sb)) {
 966                 xfs_warn(mp,
 967                   "Super block does not support project and group quota together");
 968                 return -EINVAL;
 969         }
 970
 971         return 0;
 972 }
 973
 974 static int
 975 xfs_init_percpu_counters(
 976         struct xfs_mount        *mp)
 977 {
 978         int             error;
 979
 980         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
 981         if (error)
 982                 return -ENOMEM;
 983
 984         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
 985         if (error)
 986                 goto free_icount;
 987
 988         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
 989         if (error)
 990                 goto free_ifree;
 991
 992         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
 993         if (error)
 994                 goto free_fdblocks;
 995
 996         return 0;
 997
 998 free_fdblocks:
 999         percpu_counter_destroy(&mp->m_fdblocks);
1000 free_ifree:
1001         percpu_counter_destroy(&mp->m_ifree);
1002 free_icount:
1003         percpu_counter_destroy(&mp->m_icount);
1004         return -ENOMEM;
1005 }
1006
1007 void
1008 xfs_reinit_percpu_counters(
1009         struct xfs_mount        *mp)
1010 {
1011         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1012         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1013         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1014 }
1015
1016 static void
1017 xfs_destroy_percpu_counters(
1018         struct xfs_mount        *mp)
1019 {
1020         percpu_counter_destroy(&mp->m_icount);
1021         percpu_counter_destroy(&mp->m_ifree);
1022         percpu_counter_destroy(&mp->m_fdblocks);
1023         ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1024                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1025         percpu_counter_destroy(&mp->m_delalloc_blks);
1026 }
1027
1028 static void
1029 xfs_fs_put_super(
1030         struct super_block      *sb)
1031 {
1032         struct xfs_mount        *mp = XFS_M(sb);
1033
1034         /* if ->fill_super failed, we have no mount to tear down */
1035         if (!sb->s_fs_info)
1036                 return;
1037
1038         xfs_notice(mp, "Unmounting Filesystem");
1039         xfs_filestream_unmount(mp);
1040         xfs_unmountfs(mp);
1041
1042         xfs_freesb(mp);
1043         free_percpu(mp->m_stats.xs_stats);
1044         xfs_destroy_percpu_counters(mp);
1045         xfs_destroy_mount_workqueues(mp);
1046         xfs_close_devices(mp);
1047
1048         sb->s_fs_info = NULL;
1049         xfs_mount_free(mp);
1050 }
1051
1052 static long
1053 xfs_fs_nr_cached_objects(
1054         struct super_block      *sb,
1055         struct shrink_control   *sc)
1056 {
1057         /* Paranoia: catch incorrect calls during mount setup or teardown */
1058         if (WARN_ON_ONCE(!sb->s_fs_info))
1059                 return 0;
1060         return xfs_reclaim_inodes_count(XFS_M(sb));
1061 }
1062
1063 static long
1064 xfs_fs_free_cached_objects(
1065         struct super_block      *sb,
1066         struct shrink_control   *sc)
1067 {
1068         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1069 }
1070
1071 static const struct super_operations xfs_super_operations = {
1072         .alloc_inode            = xfs_fs_alloc_inode,
1073         .destroy_inode          = xfs_fs_destroy_inode,
1074         .dirty_inode            = xfs_fs_dirty_inode,
1075         .drop_inode             = xfs_fs_drop_inode,
1076         .put_super              = xfs_fs_put_super,
1077         .sync_fs                = xfs_fs_sync_fs,
1078         .freeze_fs              = xfs_fs_freeze,
1079         .unfreeze_fs            = xfs_fs_unfreeze,
1080         .statfs                 = xfs_fs_statfs,
1081         .show_options           = xfs_fs_show_options,
1082         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1083         .free_cached_objects    = xfs_fs_free_cached_objects,
1084 };
1085
1086 static int
1087 suffix_kstrtoint(
1088         const char      *s,
1089         unsigned int    base,
1090         int             *res)
1091 {
1092         int             last, shift_left_factor = 0, _res;
1093         char            *value;
1094         int             ret = 0;
1095
1096         value = kstrdup(s, GFP_KERNEL);
1097         if (!value)
1098                 return -ENOMEM;
1099
1100         last = strlen(value) - 1;
1101         if (value[last] == 'K' || value[last] == 'k') {
1102                 shift_left_factor = 10;
1103                 value[last] = '\0';
1104         }
1105         if (value[last] == 'M' || value[last] == 'm') {
1106                 shift_left_factor = 20;
1107                 value[last] = '\0';
1108         }
1109         if (value[last] == 'G' || value[last] == 'g') {
1110                 shift_left_factor = 30;
1111                 value[last] = '\0';
1112         }
1113
1114         if (kstrtoint(value, base, &_res))
1115                 ret = -EINVAL;
1116         kfree(value);
1117         *res = _res << shift_left_factor;
1118         return ret;
1119 }
1120
1121 static inline void
1122 xfs_fs_warn_deprecated(
1123         struct fs_context       *fc,
1124         struct fs_parameter     *param,
1125         uint64_t                flag,
1126         bool                    value)
1127 {
1128         /* Don't print the warning if reconfiguring and current mount point
1129          * already had the flag set
1130          */
1131         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1132                         !!(XFS_M(fc->root->d_sb)->m_flags & flag) == value)
1133                 return;
1134         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1135 }
1136
1137 /*
1138  * Set mount state from a mount option.
1139  *
1140  * NOTE: mp->m_super is NULL here!
1141  */
1142 static int
1143 xfs_fs_parse_param(
1144         struct fs_context       *fc,
1145         struct fs_parameter     *param)
1146 {
1147         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1148         struct fs_parse_result  result;
1149         int                     size = 0;
1150         int                     opt;
1151
1152         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1153         if (opt < 0)
1154                 return opt;
1155
1156         switch (opt) {
1157         case Opt_logbufs:
1158                 parsing_mp->m_logbufs = result.uint_32;
1159                 return 0;
1160         case Opt_logbsize:
1161                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1162                         return -EINVAL;
1163                 return 0;
1164         case Opt_logdev:
1165                 kfree(parsing_mp->m_logname);
1166                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1167                 if (!parsing_mp->m_logname)
1168                         return -ENOMEM;
1169                 return 0;
1170         case Opt_rtdev:
1171                 kfree(parsing_mp->m_rtname);
1172                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1173                 if (!parsing_mp->m_rtname)
1174                         return -ENOMEM;
1175                 return 0;
1176         case Opt_allocsize:
1177                 if (suffix_kstrtoint(param->string, 10, &size))
1178                         return -EINVAL;
1179                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1180                 parsing_mp->m_flags |= XFS_MOUNT_ALLOCSIZE;
1181                 return 0;
1182         case Opt_grpid:
1183         case Opt_bsdgroups:
1184                 parsing_mp->m_flags |= XFS_MOUNT_GRPID;
1185                 return 0;
1186         case Opt_nogrpid:
1187         case Opt_sysvgroups:
1188                 parsing_mp->m_flags &= ~XFS_MOUNT_GRPID;
1189                 return 0;
1190         case Opt_wsync:
1191                 parsing_mp->m_flags |= XFS_MOUNT_WSYNC;
1192                 return 0;
1193         case Opt_norecovery:
1194                 parsing_mp->m_flags |= XFS_MOUNT_NORECOVERY;
1195                 return 0;
1196         case Opt_noalign:
1197                 parsing_mp->m_flags |= XFS_MOUNT_NOALIGN;
1198                 return 0;
1199         case Opt_swalloc:
1200                 parsing_mp->m_flags |= XFS_MOUNT_SWALLOC;
1201                 return 0;
1202         case Opt_sunit:
1203                 parsing_mp->m_dalign = result.uint_32;
1204                 return 0;
1205         case Opt_swidth:
1206                 parsing_mp->m_swidth = result.uint_32;
1207                 return 0;
1208         case Opt_inode32:
1209                 parsing_mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1210                 return 0;
1211         case Opt_inode64:
1212                 parsing_mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1213                 return 0;
1214         case Opt_nouuid:
1215                 parsing_mp->m_flags |= XFS_MOUNT_NOUUID;
1216                 return 0;
1217         case Opt_largeio:
1218                 parsing_mp->m_flags |= XFS_MOUNT_LARGEIO;
1219                 return 0;
1220         case Opt_nolargeio:
1221                 parsing_mp->m_flags &= ~XFS_MOUNT_LARGEIO;
1222                 return 0;
1223         case Opt_filestreams:
1224                 parsing_mp->m_flags |= XFS_MOUNT_FILESTREAMS;
1225                 return 0;
1226         case Opt_noquota:
1227                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1228                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1229                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
1230                 return 0;
1231         case Opt_quota:
1232         case Opt_uquota:
1233         case Opt_usrquota:
1234                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
1235                                  XFS_UQUOTA_ENFD);
1236                 return 0;
1237         case Opt_qnoenforce:
1238         case Opt_uqnoenforce:
1239                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
1240                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1241                 return 0;
1242         case Opt_pquota:
1243         case Opt_prjquota:
1244                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
1245                                  XFS_PQUOTA_ENFD);
1246                 return 0;
1247         case Opt_pqnoenforce:
1248                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
1249                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1250                 return 0;
1251         case Opt_gquota:
1252         case Opt_grpquota:
1253                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
1254                                  XFS_GQUOTA_ENFD);
1255                 return 0;
1256         case Opt_gqnoenforce:
1257                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
1258                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1259                 return 0;
1260         case Opt_discard:
1261                 parsing_mp->m_flags |= XFS_MOUNT_DISCARD;
1262                 return 0;
1263         case Opt_nodiscard:
1264                 parsing_mp->m_flags &= ~XFS_MOUNT_DISCARD;
1265                 return 0;
1266 #ifdef CONFIG_FS_DAX
1267         case Opt_dax:
1268                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1269                 return 0;
1270         case Opt_dax_enum:
1271                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1272                 return 0;
1273 #endif
1274         /* Following mount options will be removed in September 2025 */
1275         case Opt_ikeep:
1276                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_IKEEP, true);
1277                 parsing_mp->m_flags |= XFS_MOUNT_IKEEP;
1278                 return 0;
1279         case Opt_noikeep:
1280                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_IKEEP, false);
1281                 parsing_mp->m_flags &= ~XFS_MOUNT_IKEEP;
1282                 return 0;
1283         case Opt_attr2:
1284                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_ATTR2, true);
1285                 parsing_mp->m_flags |= XFS_MOUNT_ATTR2;
1286                 return 0;
1287         case Opt_noattr2:
1288                 xfs_fs_warn_deprecated(fc, param, XFS_MOUNT_NOATTR2, true);
1289                 parsing_mp->m_flags &= ~XFS_MOUNT_ATTR2;
1290                 parsing_mp->m_flags |= XFS_MOUNT_NOATTR2;
1291                 return 0;
1292         default:
1293                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1294                 return -EINVAL;
1295         }
1296
1297         return 0;
1298 }
1299
1300 static int
1301 xfs_fs_validate_params(
1302         struct xfs_mount        *mp)
1303 {
1304         /*
1305          * no recovery flag requires a read-only mount
1306          */
1307         if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
1308             !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1309                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1310                 return -EINVAL;
1311         }
1312
1313         if ((mp->m_flags & XFS_MOUNT_NOALIGN) &&
1314             (mp->m_dalign || mp->m_swidth)) {
1315                 xfs_warn(mp,
1316         "sunit and swidth options incompatible with the noalign option");
1317                 return -EINVAL;
1318         }
1319
1320         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1321                 xfs_warn(mp, "quota support not available in this kernel.");
1322                 return -EINVAL;
1323         }
1324
1325         if ((mp->m_dalign && !mp->m_swidth) ||
1326             (!mp->m_dalign && mp->m_swidth)) {
1327                 xfs_warn(mp, "sunit and swidth must be specified together");
1328                 return -EINVAL;
1329         }
1330
1331         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1332                 xfs_warn(mp,
1333         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1334                         mp->m_swidth, mp->m_dalign);
1335                 return -EINVAL;
1336         }
1337
1338         if (mp->m_logbufs != -1 &&
1339             mp->m_logbufs != 0 &&
1340             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1341              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1342                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1343                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1344                 return -EINVAL;
1345         }
1346
1347         if (mp->m_logbsize != -1 &&
1348             mp->m_logbsize !=  0 &&
1349             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1350              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1351              !is_power_of_2(mp->m_logbsize))) {
1352                 xfs_warn(mp,
1353                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1354                         mp->m_logbsize);
1355                 return -EINVAL;
1356         }
1357
1358         if ((mp->m_flags & XFS_MOUNT_ALLOCSIZE) &&
1359             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1360              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1361                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1362                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1363                 return -EINVAL;
1364         }
1365
1366         return 0;
1367 }
1368
1369 static int
1370 xfs_fs_fill_super(
1371         struct super_block      *sb,
1372         struct fs_context       *fc)
1373 {
1374         struct xfs_mount        *mp = sb->s_fs_info;
1375         struct inode            *root;
1376         int                     flags = 0, error;
1377
1378         mp->m_super = sb;
1379
1380         error = xfs_fs_validate_params(mp);
1381         if (error)
1382                 goto out_free_names;
1383
1384         sb_min_blocksize(sb, BBSIZE);
1385         sb->s_xattr = xfs_xattr_handlers;
1386         sb->s_export_op = &xfs_export_operations;
1387 #ifdef CONFIG_XFS_QUOTA
1388         sb->s_qcop = &xfs_quotactl_operations;
1389         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1390 #endif
1391         sb->s_op = &xfs_super_operations;
1392
1393         /*
1394          * Delay mount work if the debug hook is set. This is debug
1395          * instrumention to coordinate simulation of xfs mount failures with
1396          * VFS superblock operations
1397          */
1398         if (xfs_globals.mount_delay) {
1399                 xfs_notice(mp, "Delaying mount for %d seconds.",
1400                         xfs_globals.mount_delay);
1401                 msleep(xfs_globals.mount_delay * 1000);
1402         }
1403
1404         if (fc->sb_flags & SB_SILENT)
1405                 flags |= XFS_MFSI_QUIET;
1406
1407         error = xfs_open_devices(mp);
1408         if (error)
1409                 goto out_free_names;
1410
1411         error = xfs_init_mount_workqueues(mp);
1412         if (error)
1413                 goto out_close_devices;
1414
1415         error = xfs_init_percpu_counters(mp);
1416         if (error)
1417                 goto out_destroy_workqueues;
1418
1419         /* Allocate stats memory before we do operations that might use it */
1420         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1421         if (!mp->m_stats.xs_stats) {
1422                 error = -ENOMEM;
1423                 goto out_destroy_counters;
1424         }
1425
1426         error = xfs_readsb(mp, flags);
1427         if (error)
1428                 goto out_free_stats;
1429
1430         error = xfs_finish_flags(mp);
1431         if (error)
1432                 goto out_free_sb;
1433
1434         error = xfs_setup_devices(mp);
1435         if (error)
1436                 goto out_free_sb;
1437
1438         /* V4 support is undergoing deprecation. */
1439         if (!xfs_sb_version_hascrc(&mp->m_sb)) {
1440 #ifdef CONFIG_XFS_SUPPORT_V4
1441                 xfs_warn_once(mp,
1442         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1443 #else
1444                 xfs_warn(mp,
1445         "Deprecated V4 format (crc=0) not supported by kernel.");
1446                 error = -EINVAL;
1447                 goto out_free_sb;
1448 #endif
1449         }
1450
1451         /* Filesystem claims it needs repair, so refuse the mount. */
1452         if (xfs_sb_version_needsrepair(&mp->m_sb)) {
1453                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1454                 error = -EFSCORRUPTED;
1455                 goto out_free_sb;
1456         }
1457
1458         /*
1459          * Don't touch the filesystem if a user tool thinks it owns the primary
1460          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1461          * we don't check them at all.
1462          */
1463         if (mp->m_sb.sb_inprogress) {
1464                 xfs_warn(mp, "Offline file system operation in progress!");
1465                 error = -EFSCORRUPTED;
1466                 goto out_free_sb;
1467         }
1468
1469         /*
1470          * Until this is fixed only page-sized or smaller data blocks work.
1471          */
1472         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1473                 xfs_warn(mp,
1474                 "File system with blocksize %d bytes. "
1475                 "Only pagesize (%ld) or less will currently work.",
1476                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1477                 error = -ENOSYS;
1478                 goto out_free_sb;
1479         }
1480
1481         /* Ensure this filesystem fits in the page cache limits */
1482         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1483             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1484                 xfs_warn(mp,
1485                 "file system too large to be mounted on this system.");
1486                 error = -EFBIG;
1487                 goto out_free_sb;
1488         }
1489
1490         /*
1491          * XFS block mappings use 54 bits to store the logical block offset.
1492          * This should suffice to handle the maximum file size that the VFS
1493          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1494          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1495          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1496          * to check this assertion.
1497          *
1498          * Avoid integer overflow by comparing the maximum bmbt offset to the
1499          * maximum pagecache offset in units of fs blocks.
1500          */
1501         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1502                 xfs_warn(mp,
1503 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1504                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1505                          XFS_MAX_FILEOFF);
1506                 error = -EINVAL;
1507                 goto out_free_sb;
1508         }
1509
1510         error = xfs_filestream_mount(mp);
1511         if (error)
1512                 goto out_free_sb;
1513
1514         /*
1515          * we must configure the block size in the superblock before we run the
1516          * full mount process as the mount process can lookup and cache inodes.
1517          */
1518         sb->s_magic = XFS_SUPER_MAGIC;
1519         sb->s_blocksize = mp->m_sb.sb_blocksize;
1520         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1521         sb->s_maxbytes = MAX_LFS_FILESIZE;
1522         sb->s_max_links = XFS_MAXLINK;
1523         sb->s_time_gran = 1;
1524         if (xfs_sb_version_hasbigtime(&mp->m_sb)) {
1525                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1526                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1527         } else {
1528                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1529                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1530         }
1531         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1532         sb->s_iflags |= SB_I_CGROUPWB;
1533
1534         set_posix_acl_flag(sb);
1535
1536         /* version 5 superblocks support inode version counters. */
1537         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1538                 sb->s_flags |= SB_I_VERSION;
1539
1540         if (xfs_sb_version_hasbigtime(&mp->m_sb))
1541                 xfs_warn(mp,
1542  "EXPERIMENTAL big timestamp feature in use. Use at your own risk!");
1543
1544         if (mp->m_flags & XFS_MOUNT_DAX_ALWAYS) {
1545                 bool rtdev_is_dax = false, datadev_is_dax;
1546
1547                 xfs_warn(mp,
1548                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1549
1550                 datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1551                         sb->s_blocksize);
1552                 if (mp->m_rtdev_targp)
1553                         rtdev_is_dax = bdev_dax_supported(
1554                                 mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1555                 if (!rtdev_is_dax && !datadev_is_dax) {
1556                         xfs_alert(mp,
1557                         "DAX unsupported by block device. Turning off DAX.");
1558                         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
1559                 }
1560                 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1561                         xfs_alert(mp,
1562                 "DAX and reflink cannot be used together!");
1563                         error = -EINVAL;
1564                         goto out_filestream_unmount;
1565                 }
1566         }
1567
1568         if (mp->m_flags & XFS_MOUNT_DISCARD) {
1569                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1570
1571                 if (!blk_queue_discard(q)) {
1572                         xfs_warn(mp, "mounting with \"discard\" option, but "
1573                                         "the device does not support discard");
1574                         mp->m_flags &= ~XFS_MOUNT_DISCARD;
1575                 }
1576         }
1577
1578         if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1579                 if (mp->m_sb.sb_rblocks) {
1580                         xfs_alert(mp,
1581         "reflink not compatible with realtime device!");
1582                         error = -EINVAL;
1583                         goto out_filestream_unmount;
1584                 }
1585
1586                 if (xfs_globals.always_cow) {
1587                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1588                         mp->m_always_cow = true;
1589                 }
1590         }
1591
1592         if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1593                 xfs_alert(mp,
1594         "reverse mapping btree not compatible with realtime device!");
1595                 error = -EINVAL;
1596                 goto out_filestream_unmount;
1597         }
1598
1599         if (xfs_sb_version_hasinobtcounts(&mp->m_sb))
1600                 xfs_warn(mp,
1601  "EXPERIMENTAL inode btree counters feature in use. Use at your own risk!");
1602
1603         error = xfs_mountfs(mp);
1604         if (error)
1605                 goto out_filestream_unmount;
1606
1607         root = igrab(VFS_I(mp->m_rootip));
1608         if (!root) {
1609                 error = -ENOENT;
1610                 goto out_unmount;
1611         }
1612         sb->s_root = d_make_root(root);
1613         if (!sb->s_root) {
1614                 error = -ENOMEM;
1615                 goto out_unmount;
1616         }
1617
1618         return 0;
1619
1620  out_filestream_unmount:
1621         xfs_filestream_unmount(mp);
1622  out_free_sb:
1623         xfs_freesb(mp);
1624  out_free_stats:
1625         free_percpu(mp->m_stats.xs_stats);
1626  out_destroy_counters:
1627         xfs_destroy_percpu_counters(mp);
1628  out_destroy_workqueues:
1629         xfs_destroy_mount_workqueues(mp);
1630  out_close_devices:
1631         xfs_close_devices(mp);
1632  out_free_names:
1633         sb->s_fs_info = NULL;
1634         xfs_mount_free(mp);
1635         return error;
1636
1637  out_unmount:
1638         xfs_filestream_unmount(mp);
1639         xfs_unmountfs(mp);
1640         goto out_free_sb;
1641 }
1642
1643 static int
1644 xfs_fs_get_tree(
1645         struct fs_context       *fc)
1646 {
1647         return get_tree_bdev(fc, xfs_fs_fill_super);
1648 }
1649
1650 static int
1651 xfs_remount_rw(
1652         struct xfs_mount        *mp)
1653 {
1654         struct xfs_sb           *sbp = &mp->m_sb;
1655         int error;
1656
1657         if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1658                 xfs_warn(mp,
1659                         "ro->rw transition prohibited on norecovery mount");
1660                 return -EINVAL;
1661         }
1662
1663         if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1664             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1665                 xfs_warn(mp,
1666         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1667                         (sbp->sb_features_ro_compat &
1668                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1669                 return -EINVAL;
1670         }
1671
1672         mp->m_flags &= ~XFS_MOUNT_RDONLY;
1673
1674         /*
1675          * If this is the first remount to writeable state we might have some
1676          * superblock changes to update.
1677          */
1678         if (mp->m_update_sb) {
1679                 error = xfs_sync_sb(mp, false);
1680                 if (error) {
1681                         xfs_warn(mp, "failed to write sb changes");
1682                         return error;
1683                 }
1684                 mp->m_update_sb = false;
1685         }
1686
1687         /*
1688          * Fill out the reserve pool if it is empty. Use the stashed value if
1689          * it is non-zero, otherwise go with the default.
1690          */
1691         xfs_restore_resvblks(mp);
1692         xfs_log_work_queue(mp);
1693
1694         /* Recover any CoW blocks that never got remapped. */
1695         error = xfs_reflink_recover_cow(mp);
1696         if (error) {
1697                 xfs_err(mp,
1698                         "Error %d recovering leftover CoW allocations.", error);
1699                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1700                 return error;
1701         }
1702         xfs_blockgc_start(mp);
1703
1704         /* Create the per-AG metadata reservation pool .*/
1705         error = xfs_fs_reserve_ag_blocks(mp);
1706         if (error && error != -ENOSPC)
1707                 return error;
1708
1709         return 0;
1710 }
1711
1712 static int
1713 xfs_remount_ro(
1714         struct xfs_mount        *mp)
1715 {
1716         int error;
1717
1718         /*
1719          * Cancel background eofb scanning so it cannot race with the final
1720          * log force+buftarg wait and deadlock the remount.
1721          */
1722         xfs_blockgc_stop(mp);
1723
1724         /* Get rid of any leftover CoW reservations... */
1725         error = xfs_blockgc_free_space(mp, NULL);
1726         if (error) {
1727                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1728                 return error;
1729         }
1730
1731         /* Free the per-AG metadata reservation pool. */
1732         error = xfs_fs_unreserve_ag_blocks(mp);
1733         if (error) {
1734                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1735                 return error;
1736         }
1737
1738         /*
1739          * Before we sync the metadata, we need to free up the reserve block
1740          * pool so that the used block count in the superblock on disk is
1741          * correct at the end of the remount. Stash the current* reserve pool
1742          * size so that if we get remounted rw, we can return it to the same
1743          * size.
1744          */
1745         xfs_save_resvblks(mp);
1746
1747         xfs_log_clean(mp);
1748         mp->m_flags |= XFS_MOUNT_RDONLY;
1749
1750         return 0;
1751 }
1752
1753 /*
1754  * Logically we would return an error here to prevent users from believing
1755  * they might have changed mount options using remount which can't be changed.
1756  *
1757  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1758  * arguments in some cases so we can't blindly reject options, but have to
1759  * check for each specified option if it actually differs from the currently
1760  * set option and only reject it if that's the case.
1761  *
1762  * Until that is implemented we return success for every remount request, and
1763  * silently ignore all options that we can't actually change.
1764  */
1765 static int
1766 xfs_fs_reconfigure(
1767         struct fs_context *fc)
1768 {
1769         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1770         struct xfs_mount        *new_mp = fc->s_fs_info;
1771         xfs_sb_t                *sbp = &mp->m_sb;
1772         int                     flags = fc->sb_flags;
1773         int                     error;
1774
1775         /* version 5 superblocks always support version counters. */
1776         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1777                 fc->sb_flags |= SB_I_VERSION;
1778
1779         error = xfs_fs_validate_params(new_mp);
1780         if (error)
1781                 return error;
1782
1783         sync_filesystem(mp->m_super);
1784
1785         /* inode32 -> inode64 */
1786         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1787             !(new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1788                 mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1789                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1790         }
1791
1792         /* inode64 -> inode32 */
1793         if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS) &&
1794             (new_mp->m_flags & XFS_MOUNT_SMALL_INUMS)) {
1795                 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1796                 mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1797         }
1798
1799         /* ro -> rw */
1800         if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(flags & SB_RDONLY)) {
1801                 error = xfs_remount_rw(mp);
1802                 if (error)
1803                         return error;
1804         }
1805
1806         /* rw -> ro */
1807         if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (flags & SB_RDONLY)) {
1808                 error = xfs_remount_ro(mp);
1809                 if (error)
1810                         return error;
1811         }
1812
1813         return 0;
1814 }
1815
1816 static void xfs_fs_free(
1817         struct fs_context       *fc)
1818 {
1819         struct xfs_mount        *mp = fc->s_fs_info;
1820
1821         /*
1822          * mp is stored in the fs_context when it is initialized.
1823          * mp is transferred to the superblock on a successful mount,
1824          * but if an error occurs before the transfer we have to free
1825          * it here.
1826          */
1827         if (mp)
1828                 xfs_mount_free(mp);
1829 }
1830
1831 static const struct fs_context_operations xfs_context_ops = {
1832         .parse_param = xfs_fs_parse_param,
1833         .get_tree    = xfs_fs_get_tree,
1834         .reconfigure = xfs_fs_reconfigure,
1835         .free        = xfs_fs_free,
1836 };
1837
1838 static int xfs_init_fs_context(
1839         struct fs_context       *fc)
1840 {
1841         struct xfs_mount        *mp;
1842
1843         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1844         if (!mp)
1845                 return -ENOMEM;
1846
1847         spin_lock_init(&mp->m_sb_lock);
1848         spin_lock_init(&mp->m_agirotor_lock);
1849         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1850         spin_lock_init(&mp->m_perag_lock);
1851         mutex_init(&mp->m_growlock);
1852         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1853         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1854         mp->m_kobj.kobject.kset = xfs_kset;
1855         /*
1856          * We don't create the finobt per-ag space reservation until after log
1857          * recovery, so we must set this to true so that an ifree transaction
1858          * started during log recovery will not depend on space reservations
1859          * for finobt expansion.
1860          */
1861         mp->m_finobt_nores = true;
1862
1863         /*
1864          * These can be overridden by the mount option parsing.
1865          */
1866         mp->m_logbufs = -1;
1867         mp->m_logbsize = -1;
1868         mp->m_allocsize_log = 16; /* 64k */
1869
1870         /*
1871          * Copy binary VFS mount flags we are interested in.
1872          */
1873         if (fc->sb_flags & SB_RDONLY)
1874                 mp->m_flags |= XFS_MOUNT_RDONLY;
1875         if (fc->sb_flags & SB_DIRSYNC)
1876                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
1877         if (fc->sb_flags & SB_SYNCHRONOUS)
1878                 mp->m_flags |= XFS_MOUNT_WSYNC;
1879
1880         fc->s_fs_info = mp;
1881         fc->ops = &xfs_context_ops;
1882
1883         return 0;
1884 }
1885
1886 static struct file_system_type xfs_fs_type = {
1887         .owner                  = THIS_MODULE,
1888         .name                   = "xfs",
1889         .init_fs_context        = xfs_init_fs_context,
1890         .parameters             = xfs_fs_parameters,
1891         .kill_sb                = kill_block_super,
1892         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1893 };
1894 MODULE_ALIAS_FS("xfs");
1895
1896 STATIC int __init
1897 xfs_init_zones(void)
1898 {
1899         xfs_log_ticket_zone = kmem_cache_create("xfs_log_ticket",
1900                                                 sizeof(struct xlog_ticket),
1901                                                 0, 0, NULL);
1902         if (!xfs_log_ticket_zone)
1903                 goto out;
1904
1905         xfs_bmap_free_item_zone = kmem_cache_create("xfs_bmap_free_item",
1906                                         sizeof(struct xfs_extent_free_item),
1907                                         0, 0, NULL);
1908         if (!xfs_bmap_free_item_zone)
1909                 goto out_destroy_log_ticket_zone;
1910
1911         xfs_btree_cur_zone = kmem_cache_create("xfs_btree_cur",
1912                                                sizeof(struct xfs_btree_cur),
1913                                                0, 0, NULL);
1914         if (!xfs_btree_cur_zone)
1915                 goto out_destroy_bmap_free_item_zone;
1916
1917         xfs_da_state_zone = kmem_cache_create("xfs_da_state",
1918                                               sizeof(struct xfs_da_state),
1919                                               0, 0, NULL);
1920         if (!xfs_da_state_zone)
1921                 goto out_destroy_btree_cur_zone;
1922
1923         xfs_ifork_zone = kmem_cache_create("xfs_ifork",
1924                                            sizeof(struct xfs_ifork),
1925                                            0, 0, NULL);
1926         if (!xfs_ifork_zone)
1927                 goto out_destroy_da_state_zone;
1928
1929         xfs_trans_zone = kmem_cache_create("xfs_trans",
1930                                            sizeof(struct xfs_trans),
1931                                            0, 0, NULL);
1932         if (!xfs_trans_zone)
1933                 goto out_destroy_ifork_zone;
1934
1935
1936         /*
1937          * The size of the zone allocated buf log item is the maximum
1938          * size possible under XFS.  This wastes a little bit of memory,
1939          * but it is much faster.
1940          */
1941         xfs_buf_item_zone = kmem_cache_create("xfs_buf_item",
1942                                               sizeof(struct xfs_buf_log_item),
1943                                               0, 0, NULL);
1944         if (!xfs_buf_item_zone)
1945                 goto out_destroy_trans_zone;
1946
1947         xfs_efd_zone = kmem_cache_create("xfs_efd_item",
1948                                         (sizeof(struct xfs_efd_log_item) +
1949                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
1950                                         sizeof(struct xfs_extent)),
1951                                         0, 0, NULL);
1952         if (!xfs_efd_zone)
1953                 goto out_destroy_buf_item_zone;
1954
1955         xfs_efi_zone = kmem_cache_create("xfs_efi_item",
1956                                          (sizeof(struct xfs_efi_log_item) +
1957                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
1958                                          sizeof(struct xfs_extent)),
1959                                          0, 0, NULL);
1960         if (!xfs_efi_zone)
1961                 goto out_destroy_efd_zone;
1962
1963         xfs_inode_zone = kmem_cache_create("xfs_inode",
1964                                            sizeof(struct xfs_inode), 0,
1965                                            (SLAB_HWCACHE_ALIGN |
1966                                             SLAB_RECLAIM_ACCOUNT |
1967                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1968                                            xfs_fs_inode_init_once);
1969         if (!xfs_inode_zone)
1970                 goto out_destroy_efi_zone;
1971
1972         xfs_ili_zone = kmem_cache_create("xfs_ili",
1973                                          sizeof(struct xfs_inode_log_item), 0,
1974                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1975                                          NULL);
1976         if (!xfs_ili_zone)
1977                 goto out_destroy_inode_zone;
1978
1979         xfs_icreate_zone = kmem_cache_create("xfs_icr",
1980                                              sizeof(struct xfs_icreate_item),
1981                                              0, 0, NULL);
1982         if (!xfs_icreate_zone)
1983                 goto out_destroy_ili_zone;
1984
1985         xfs_rud_zone = kmem_cache_create("xfs_rud_item",
1986                                          sizeof(struct xfs_rud_log_item),
1987                                          0, 0, NULL);
1988         if (!xfs_rud_zone)
1989                 goto out_destroy_icreate_zone;
1990
1991         xfs_rui_zone = kmem_cache_create("xfs_rui_item",
1992                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1993                         0, 0, NULL);
1994         if (!xfs_rui_zone)
1995                 goto out_destroy_rud_zone;
1996
1997         xfs_cud_zone = kmem_cache_create("xfs_cud_item",
1998                                          sizeof(struct xfs_cud_log_item),
1999                                          0, 0, NULL);
2000         if (!xfs_cud_zone)
2001                 goto out_destroy_rui_zone;
2002
2003         xfs_cui_zone = kmem_cache_create("xfs_cui_item",
2004                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2005                         0, 0, NULL);
2006         if (!xfs_cui_zone)
2007                 goto out_destroy_cud_zone;
2008
2009         xfs_bud_zone = kmem_cache_create("xfs_bud_item",
2010                                          sizeof(struct xfs_bud_log_item),
2011                                          0, 0, NULL);
2012         if (!xfs_bud_zone)
2013                 goto out_destroy_cui_zone;
2014
2015         xfs_bui_zone = kmem_cache_create("xfs_bui_item",
2016                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2017                         0, 0, NULL);
2018         if (!xfs_bui_zone)
2019                 goto out_destroy_bud_zone;
2020
2021         return 0;
2022
2023  out_destroy_bud_zone:
2024         kmem_cache_destroy(xfs_bud_zone);
2025  out_destroy_cui_zone:
2026         kmem_cache_destroy(xfs_cui_zone);
2027  out_destroy_cud_zone:
2028         kmem_cache_destroy(xfs_cud_zone);
2029  out_destroy_rui_zone:
2030         kmem_cache_destroy(xfs_rui_zone);
2031  out_destroy_rud_zone:
2032         kmem_cache_destroy(xfs_rud_zone);
2033  out_destroy_icreate_zone:
2034         kmem_cache_destroy(xfs_icreate_zone);
2035  out_destroy_ili_zone:
2036         kmem_cache_destroy(xfs_ili_zone);
2037  out_destroy_inode_zone:
2038         kmem_cache_destroy(xfs_inode_zone);
2039  out_destroy_efi_zone:
2040         kmem_cache_destroy(xfs_efi_zone);
2041  out_destroy_efd_zone:
2042         kmem_cache_destroy(xfs_efd_zone);
2043  out_destroy_buf_item_zone:
2044         kmem_cache_destroy(xfs_buf_item_zone);
2045  out_destroy_trans_zone:
2046         kmem_cache_destroy(xfs_trans_zone);
2047  out_destroy_ifork_zone:
2048         kmem_cache_destroy(xfs_ifork_zone);
2049  out_destroy_da_state_zone:
2050         kmem_cache_destroy(xfs_da_state_zone);
2051  out_destroy_btree_cur_zone:
2052         kmem_cache_destroy(xfs_btree_cur_zone);
2053  out_destroy_bmap_free_item_zone:
2054         kmem_cache_destroy(xfs_bmap_free_item_zone);
2055  out_destroy_log_ticket_zone:
2056         kmem_cache_destroy(xfs_log_ticket_zone);
2057  out:
2058         return -ENOMEM;
2059 }
2060
2061 STATIC void
2062 xfs_destroy_zones(void)
2063 {
2064         /*
2065          * Make sure all delayed rcu free are flushed before we
2066          * destroy caches.
2067          */
2068         rcu_barrier();
2069         kmem_cache_destroy(xfs_bui_zone);
2070         kmem_cache_destroy(xfs_bud_zone);
2071         kmem_cache_destroy(xfs_cui_zone);
2072         kmem_cache_destroy(xfs_cud_zone);
2073         kmem_cache_destroy(xfs_rui_zone);
2074         kmem_cache_destroy(xfs_rud_zone);
2075         kmem_cache_destroy(xfs_icreate_zone);
2076         kmem_cache_destroy(xfs_ili_zone);
2077         kmem_cache_destroy(xfs_inode_zone);
2078         kmem_cache_destroy(xfs_efi_zone);
2079         kmem_cache_destroy(xfs_efd_zone);
2080         kmem_cache_destroy(xfs_buf_item_zone);
2081         kmem_cache_destroy(xfs_trans_zone);
2082         kmem_cache_destroy(xfs_ifork_zone);
2083         kmem_cache_destroy(xfs_da_state_zone);
2084         kmem_cache_destroy(xfs_btree_cur_zone);
2085         kmem_cache_destroy(xfs_bmap_free_item_zone);
2086         kmem_cache_destroy(xfs_log_ticket_zone);
2087 }
2088
2089 STATIC int __init
2090 xfs_init_workqueues(void)
2091 {
2092         /*
2093          * The allocation workqueue can be used in memory reclaim situations
2094          * (writepage path), and parallelism is only limited by the number of
2095          * AGs in all the filesystems mounted. Hence use the default large
2096          * max_active value for this workqueue.
2097          */
2098         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2099                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2100         if (!xfs_alloc_wq)
2101                 return -ENOMEM;
2102
2103         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2104                         0);
2105         if (!xfs_discard_wq)
2106                 goto out_free_alloc_wq;
2107
2108         return 0;
2109 out_free_alloc_wq:
2110         destroy_workqueue(xfs_alloc_wq);
2111         return -ENOMEM;
2112 }
2113
2114 STATIC void
2115 xfs_destroy_workqueues(void)
2116 {
2117         destroy_workqueue(xfs_discard_wq);
2118         destroy_workqueue(xfs_alloc_wq);
2119 }
2120
2121 STATIC int __init
2122 init_xfs_fs(void)
2123 {
2124         int                     error;
2125
2126         xfs_check_ondisk_structs();
2127
2128         printk(KERN_INFO XFS_VERSION_STRING " with "
2129                          XFS_BUILD_OPTIONS " enabled\n");
2130
2131         xfs_dir_startup();
2132
2133         error = xfs_init_zones();
2134         if (error)
2135                 goto out;
2136
2137         error = xfs_init_workqueues();
2138         if (error)
2139                 goto out_destroy_zones;
2140
2141         error = xfs_mru_cache_init();
2142         if (error)
2143                 goto out_destroy_wq;
2144
2145         error = xfs_buf_init();
2146         if (error)
2147                 goto out_mru_cache_uninit;
2148
2149         error = xfs_init_procfs();
2150         if (error)
2151                 goto out_buf_terminate;
2152
2153         error = xfs_sysctl_register();
2154         if (error)
2155                 goto out_cleanup_procfs;
2156
2157         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2158         if (!xfs_kset) {
2159                 error = -ENOMEM;
2160                 goto out_sysctl_unregister;
2161         }
2162
2163         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2164
2165         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2166         if (!xfsstats.xs_stats) {
2167                 error = -ENOMEM;
2168                 goto out_kset_unregister;
2169         }
2170
2171         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2172                                "stats");
2173         if (error)
2174                 goto out_free_stats;
2175
2176 #ifdef DEBUG
2177         xfs_dbg_kobj.kobject.kset = xfs_kset;
2178         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2179         if (error)
2180                 goto out_remove_stats_kobj;
2181 #endif
2182
2183         error = xfs_qm_init();
2184         if (error)
2185                 goto out_remove_dbg_kobj;
2186
2187         error = register_filesystem(&xfs_fs_type);
2188         if (error)
2189                 goto out_qm_exit;
2190         return 0;
2191
2192  out_qm_exit:
2193         xfs_qm_exit();
2194  out_remove_dbg_kobj:
2195 #ifdef DEBUG
2196         xfs_sysfs_del(&xfs_dbg_kobj);
2197  out_remove_stats_kobj:
2198 #endif
2199         xfs_sysfs_del(&xfsstats.xs_kobj);
2200  out_free_stats:
2201         free_percpu(xfsstats.xs_stats);
2202  out_kset_unregister:
2203         kset_unregister(xfs_kset);
2204  out_sysctl_unregister:
2205         xfs_sysctl_unregister();
2206  out_cleanup_procfs:
2207         xfs_cleanup_procfs();
2208  out_buf_terminate:
2209         xfs_buf_terminate();
2210  out_mru_cache_uninit:
2211         xfs_mru_cache_uninit();
2212  out_destroy_wq:
2213         xfs_destroy_workqueues();
2214  out_destroy_zones:
2215         xfs_destroy_zones();
2216  out:
2217         return error;
2218 }
2219
2220 STATIC void __exit
2221 exit_xfs_fs(void)
2222 {
2223         xfs_qm_exit();
2224         unregister_filesystem(&xfs_fs_type);
2225 #ifdef DEBUG
2226         xfs_sysfs_del(&xfs_dbg_kobj);
2227 #endif
2228         xfs_sysfs_del(&xfsstats.xs_kobj);
2229         free_percpu(xfsstats.xs_stats);
2230         kset_unregister(xfs_kset);
2231         xfs_sysctl_unregister();
2232         xfs_cleanup_procfs();
2233         xfs_buf_terminate();
2234         xfs_mru_cache_uninit();
2235         xfs_destroy_workqueues();
2236         xfs_destroy_zones();
2237         xfs_uuid_table_free();
2238 }
2239
2240 module_init(init_xfs_fs);
2241 module_exit(exit_xfs_fs);
2242
2243 MODULE_AUTHOR("Silicon Graphics, Inc.");
2244 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2245 MODULE_LICENSE("GPL");